Semi-automated tree-cadastre updating and tree classification based on high-resolution aerial RGB-imagery in Melville, Australia
Tree surveys with the objective of establishing a tree cadastre or communal tree inventory is a time-consuming and expensive work.1 As cadastres are commonly acquired in laborious eld surveys and updating involves regular site inspection, the effort of keeping a cadastre up-to-date is often either too high,2 or a tree inventory is created only once or updated in a coarse temporal resolution. In the underlying study, we present a hybrid approach of merging data from different sources, to update a cadastre (shapefile) containing tree data. A classification of the four most frequent tree species in a study domain in Melville, Western Australia, was carried out. The considered tree species were Jacaranda Mimosifolia, Agonis Flexuosa, Callistemon KP Special, and Ulmus Parvifolia. The classification was performed on high-resolution airborne imagery, using Random Forests, and achieved outstanding results with an overall model accuracy of 93:44% and Cohen's of 89:93 %. This is a considerable step towards automated generation of communal tree cadastres in the contemplated geopgraphic domain. The proposed method demonstrates that (1) high-resolution aerial imagery has great potential in being a precise and efficient alternative for updating or creating communal tree cadastres, (2) updating requires minimal user interaction and can potentially be performed in a fully automated process, and (3) based on the excellent classification results, the considered tree species can now be detected and accurately mapped at scale.